Relational Learning Imitation System for Generalization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning systems can be brittle, over-specified, and prone to overfitting, requiring additional input information and failing to generalize well to new situations due to their reliance on specific training data and reinforcement learning feedback.

Innovation Solution

The development of an imitation system that learns a relational model by monitoring the behavior of an existing system, allowing it to reproduce input-output characteristics without requiring additional input data, using techniques like Markov Logic Networks and probabilistic relational models to infer relationships and generalize behavior.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If reinforcement learning techniques are employed with specific training data and feedback, then the system can learn to perform tasks intelligently, but the system becomes brittle and over-specified, requiring extra input information and failing to generalize well

Engineering Contradiction:

Improvegeneralization capabilityVSAvoidsystem brittleness

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent creates an imitation system that copies the input-output behavior of an existing system by monitoring its behavior and learning a relational model that reproduces these characteristics. This allows the new system to generalize better without being brittle, as it learns the underlying relational structure rather than memorizing specific training examples.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent transitions from traditional machine learning approaches to relational learning, changing the fundamental parameter representation from propositional to relational. This allows the system to capture relationships between objects and generalize to new situations without requiring additional input information or becoming over-specified.

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If traditional machine learning algorithms are used with propositional models, then the system can process instructions and perform tasks, but it fails to contemplate relationships between items, limiting its ability to handle new situations

Engineering Contradiction:

Improverelationship reasoning capabilityVSAvoidperformance consistency

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent introduces a new dimension to the learning problem by moving from propositional representation to relational representation. This dimensional change allows the system to explicitly model relationships between items, enabling it to reason about new situations by understanding the relational structure rather than relying on predefined categories.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent substitutes traditional propositional machine learning mechanisms with relational learning mechanisms. Instead of using propositional models that process isolated facts, the system uses relational models that capture relationships between objects, thereby improving both adaptability and reliability through relationship-based reasoning.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS8862523B2Relational learning for system imitation

Publication Date: 2014.10.14 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Technologies pertaining to learning a computer-executable imitation system that imitates behavior of an existing computer-executable system are described herein. Behavior of an existing computer-executable system can be monitored through monitoring data input to the existing computer-executable system and data output by the existing computer-executable system responsive to receipt of the input data. An imitation system that imitates the behavior of the existing system can be learned, wherein the imitation system comprises a relational model.